# Link MCP server

> Provides local, Markdown-based memory for AI agents to capture, structure, and recall preferences, decisions, and project context.

- Repository: https://github.com/gowtham0992/link
- Page: https://hostedmcp.org/servers/gowtham0992/link
- Transport: stdio
- Stars: 218
- License: MIT
- Latest version: 5.0.0

## Install

### Claude Code

```sh
claude mcp add link -- uvx link-mcp
```

### Cursor (`~/.cursor/mcp.json`)

```json
{
  "mcpServers": {
    "link": {
      "command": "uvx",
      "args": [
        "link-mcp"
      ]
    }
  }
}
```

### VS Code (`.vscode/mcp.json`)

```json
{
  "servers": {
    "link": {
      "type": "stdio",
      "command": "uvx",
      "args": [
        "link-mcp"
      ]
    }
  }
}
```

### VS Code CLI

```sh
code --add-mcp '{"name":"link","type":"stdio","command":"uvx","args":["link-mcp"]}'
```

### Claude Desktop (`claude_desktop_config.json`)

```json
{
  "mcpServers": {
    "link": {
      "command": "uvx",
      "args": [
        "link-mcp"
      ]
    }
  }
}
```

### Codex

```sh
codex mcp add link -- uvx link-mcp
```

### Codex config.toml (`~/.codex/config.toml`)

```toml
[mcp_servers.link]
command = "uvx"
args = ["link-mcp"]
```

### Gemini CLI

```sh
gemini mcp add -s user link uvx -- link-mcp
```

### Gemini settings.json (`~/.gemini/settings.json`)

```json
{
  "mcpServers": {
    "link": {
      "command": "uvx",
      "args": [
        "link-mcp"
      ]
    }
  }
}
```

## FAQ

### How is the memory stored?

Memory is stored locally as plain Markdown files that you can open, grep, or git-diff.

### Can I trust the agent not to hallucinate facts into my memory?

Yes, because there is no LLM in the write path, ingestion and recall are deterministic.

### How do I set up Link for my agents?

You can install it via Homebrew or pip, then run the setup command to wire up your agents.

### Does Link require an internet connection to function?

It is provably local; the runtime blocks outbound network code and semantic models can load offline.
